LLM Council by Felix

The LLM Council is a decision-making pattern popularised by Andrej Karpathy, who open-sourced a small “Saturday hack” app for it in late 2025. Instead of trusting a single model's answer, it sends your question to several models at once, has them anonymously review and critique each other's responses, and then asks a “chairman” model to weigh everything and write the final verdict.

Why a council beats a single answer

A lone model tends to sound equally confident whether or not it's right, and it rarely flags its own blind spots. Forcing several perspectives into the open — and making them judge each other before anything is finalised — surfaces disagreement you'd otherwise never see. The synthesis step then turns that debate into one clear recommendation instead of a wall of conflicting opinions.

What Felix's version adds

Felix's adaptation (hosted on ADPList's Notion) packages the idea as a free, ready-to-use Claude add-on, so you don't have to wire up multiple model APIs yourself. You describe the decision you're weighing, and it stands up a panel of five AI advisers — each taking a distinct angle — that deliberate and then hand you a single verdict plus a concrete next step.

When to reach for it

• Strategy calls where the “obvious” answer feels a little too clean
• Hiring, pricing, or “should we build this?” decisions with real trade-offs
• Any choice where a second, third, and fourth opinion would genuinely sharpen your thinking

It's less useful for quick factual lookups, where the overhead of a full council isn't worth it. For a weighty, ambiguous decision, though, the deliberation is exactly the point.